WO2023127262A1 - Machine learning device, exhaust gas analysis device, machine learning method, exhaust gas analysis method, machine learning program, and exhaust gas analysis program - Google Patents

Abstract

A machine learning device 6 which is to be used in an exhaust gas analysis device which irradiates exhaust gas with light and detects the light which has passed through the exhaust gas, and analyzes the exhaust gas on the basis of the detection signal thereof, said machine learning device 6 being equipped with: a training data receiving unit 61 for receiving training data which includes a reference value of a specific component concentration, which is the H₂ concentration and/or the O₂ concentration obtained by an analyzer which is different than the exhaust gas analysis device, and also includes at least one of the spectrum data obtained by irradiating the exhaust gas with light, an individual component concentration which is selected on the basis of an elemental balance equation for obtaining the specific component concentration, and a calculated value of the specific component concentration which is calculated via the elemental balance equation by using the individual component concentration; and a machine learning unit 62 which, by using the training data, causes machine learning for the relationship between the reference value of the specific component concentration and the spectrum data, the individual component concentration or the calculated value of the specific component concentration.

Description

Machine learning device, exhaust gas analyzer, machine learning method, exhaust gas analysis method, machine learning program, and exhaust gas analysis program

The present invention relates to a machine learning device, an exhaust gas analyzer, a machine learning method, an exhaust gas analysis method, a machine learning program, and an exhaust gas analysis program.

Conventionally, an FTIR analyzer using Fourier transform infrared spectroscopy (FTIR) has been used to analyze components contained in exhaust gas, as shown in Patent Document 1. With this FTIR analyzer, multiple components such as CO, _{CO2 ,} NO, _H2O , _NO2 , _C2H5OH , HCHO or _CH4 in exhaust gas can be analyzed simultaneously.

However, although the FTIR spectrometer can analyze components that absorb infrared rays, it cannot analyze components that do not absorb infrared rays. Therefore, when measuring the concentration of H ₂ that does not absorb infrared rays, a dedicated H ₂ analyzer such as a thermal conductivity gas analyzer (TCD) is required in addition to the FTIR analyzer. Moreover, when measuring the concentration of O ₂ that does not absorb infrared rays, a dedicated O ₂ analyzer such as a zirconia sensor is required in addition to the FTIR analyzer. As a result, installation space for both the FTIR analyzer and the H ₂ analyzer or O ₂ analyzer is required, resulting in an increase in the size of the exhaust gas analyzer. Such problems can occur not only in FTIR analyzers but also in other exhaust gas analyzers using light.

JP-A-2000-346801

Therefore, the present invention has been made in view of the above-described problems, and provides an exhaust gas analyzer capable of measuring H ₂ concentration or O ₂ concentration, which had to be measured using another analyzer. is the main subject.

That is, the machine learning device according to the present invention is a machine learning device used in an exhaust gas analyzer that irradiates a combustion exhaust gas with light, detects the light transmitted through the combustion exhaust gas, and analyzes the combustion exhaust gas based on the detection signal. An apparatus comprising a training data reception unit that receives training data and a machine learning unit that performs machine learning using the training data, wherein the training data reception unit is obtained by an analyzer different from the exhaust gas analyzer. Based on the reference value of the specific component concentration which is at least one of the H ₂ concentration or O ₂ concentration, the spectral data obtained by irradiating the combustion exhaust gas with light, and the element balance formula for obtaining the specific component concentration or at least one of the calculated values of the specific component concentration calculated using the individual component concentration in the element balance formula, and the machine learning unit generates specific component correlation data by machine-learning the relationship between the reference value of the specific component concentration and at least one of the spectral data, the individual component concentration, or the calculated value of the specific component concentration. It is characterized by

With such a configuration, the reference value of the specific component concentration that is at least one of the H ₂ concentration or the O ₂ concentration, the spectrum data obtained by irradiating the combustion exhaust gas with light, and the specific component concentration By machine learning the relationship between at least one of the individual component concentration selected based on the element balance formula or the calculated value of the specific component concentration calculated using the individual component concentration in the element balance formula, the machine Using a machine learning model generated by learning, spectrum data obtained by irradiating light on combustion exhaust gas, individual component concentration obtained by exhaust gas analyzer, or calculated from individual component concentration and element balance formula The specific component concentration can be calculated from at least one of the calculated values of the specific component concentration. As a result, the H ₂ concentration or O ₂ concentration, which had to be measured using another analyzer in the exhaust gas analyzer, can be measured. Particularly in exhaust gas analysis using infrared light, the H ₂ concentration or O ₂ concentration that does not absorb infrared light can be measured.

Further, in the machine learning device of the present invention, the teacher data receiving unit receives teacher data including the reference value of the specific component concentration and the spectrum data, and the machine learning unit receives the specific component concentration It is preferable that machine learning is performed on the relationship between the reference value of and the spectrum data to generate the specific component correlation data.

Furthermore, in order to accurately measure the H ₂ concentration or the O ₂ concentration, the teacher data receiving unit further receives the individual component concentration as teacher data, and the machine learning unit receives the specific It is preferable that the specific component correlation data be generated by machine learning of the relationship between the component concentration reference value, the spectrum data, and the individual component concentration.

As a specific embodiment of machine learning using the reference value of the specific component concentration, the calculated value of the specific component concentration, and the spectrum data, the error between the reference value of the specific component concentration and the calculated value of the specific component concentration is calculated. calculating the minimized error minimum value and generating, as part of the specific component correlation data, first correlation data indicating the correlation between the minimum error value and a parameter used to calculate the minimum error value; A correlation data generation unit machine-learns the relationship between the spectral data and the minimum error value, and generates second correlation data indicating the correlation between the spectral data and the minimum error value as part of the specific component correlation data. It is desirable to have a second correlation data generator that generates the second correlation data generator.

Further, in the machine learning device of the present invention, the teacher data receiving unit receives teacher data including the reference value of the concentration of the specific component and the concentration of the individual component, and the machine learning unit receives the specific component It is preferable that the specific component correlation data be generated by machine learning the relationship between the concentration reference value and the individual component concentration.

When machine-learning H ₂ correlation data as the specific component correlation data, the individual component concentration may be at least one of CO ₂ concentration, CO concentration, H ₂ O concentration, or THC concentration. Further, when machine learning is performed on O ₂ correlation data as the specific component correlation data, the individual component concentration is at least one of CO ₂ concentration, CO concentration, H ₂ O concentration, THC concentration, or NO concentration. can be considered.

In addition, when measuring total hydrocarbons (THC) with a conventional exhaust gas analyzer, the concentration of each hydrocarbon (HC) is obtained individually from the spectral data, and then weighted and added together. A two-stage calculation is performed, and an error that may occur in setting the weighting coefficient is superimposed on an error that may occur in the concentration measurement of each HC. Therefore, it is difficult to improve the measurement accuracy.
In order to improve the measurement accuracy of the THC concentration in the exhaust gas analyzer, the teacher data reception unit receives teacher data including the reference value of the THC concentration obtained by an analyzer different from the exhaust gas analyzer and the spectrum data. It is preferable that the machine learning unit machine-learns the relationship between the THC concentration reference value and the spectrum data to generate THC correlation data.

Here, the individual component concentration includes the THC concentration, and it is desirable that the THC concentration be obtained from the spectral data obtained by the exhaust gas analyzer and the THC correlation data.

Further, an exhaust gas analyzer according to the present invention is an exhaust gas analyzer for analyzing combustion exhaust gas, comprising: a light source for irradiating the combustion exhaust gas with light; a photodetector for detecting light transmitted through the combustion exhaust gas; Specific component concentration which is at least one of H ₂ concentration or O ₂ concentration in combustion exhaust gas, spectrum data obtained by irradiating the combustion exhaust gas with light, element balance formula for obtaining the specific component concentration Specific component correlation storing specific component correlation data obtained by learning a relationship between at least one of the individual component concentration selected by the method and the calculated value of the specific component concentration calculated using the individual component concentration in the element balance formula Specific component concentration for calculating the specific component concentration in the flue gas from a data storage unit, at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration, and the specific component correlation data. and a calculator.

With such a configuration, the specific component concentration, which is at least one of the H ₂ concentration or O ₂ concentration in the combustion exhaust gas, the spectrum data obtained by irradiating the combustion exhaust gas with light, and the specific component concentration are obtained. Specific component correlation data (machine learning Spectral data obtained by irradiating the combustion exhaust gas with light using a model), individual component concentrations obtained by an exhaust gas analyzer, or calculated values of specific component concentrations calculated from individual component concentrations and elemental balance equations From at least one of, the specific component concentration can be calculated. As a result, the H ₂ concentration or O ₂ concentration that would otherwise have to be measured using another analyzer can be measured in the analyzer. In particular, in exhaust gas analysis using infrared light, the H ₂ concentration or O ₂ concentration that does not absorb infrared light can be measured.

The exhaust gas analyzer of the present invention further includes a THC correlation data storage unit for storing THC correlation data obtained by learning the relationship between the THC concentration reference value obtained by an analyzer different from the exhaust gas analyzer and the spectrum data, It is desirable to further include a THC concentration calculation section for calculating the THC concentration in the combustion exhaust gas from the spectral data obtained by irradiating the combustion exhaust gas with light and the THC correlation data. With this configuration, the THC concentration in the combustion exhaust gas can be measured with high accuracy.

Further, it is preferable that the individual component concentration includes the THC concentration, and the THC concentration is calculated by the THC concentration calculating section. With this configuration, when the H ₂ concentration or the O ₂ concentration is measured using the THC concentration, the H ₂ concentration or the O ₂ concentration can be measured with high accuracy.

As a specific embodiment for measuring the H ₂ concentration or the O ₂ concentration using the reference value of the specific component concentration, the calculated value of the specific component concentration, and the spectrum data, the learned model storage unit stores the specific component concentration a first correlation data storage unit for storing first correlation data indicating the correlation between the minimum error value between the reference value of and the calculated value of the concentration of the specific component and the parameter used to calculate the minimum error value; a second correlation data storage unit that stores second correlation data indicating the correlation between the data and the minimum error value, and the specific component concentration calculation unit calculates the It is desirable to calculate the specific component concentration in the combustion exhaust gas from a minimum error value calculating section for calculating the minimum error value, and the minimum error value obtained by the minimum error value calculating section and the first correlation data. .

As a specific embodiment in which the effect of the present invention is remarkably exhibited, the combustion exhaust gas may be exhaust gas from an automobile. A so-called FTIR method is preferable.

Furthermore, the machine learning method according to the present invention is a machine learning method used in an exhaust gas analyzer that irradiates light on combustion exhaust gas, detects light that has passed through the combustion exhaust gas, and analyzes the combustion exhaust gas based on the detection signal. A method comprising a teacher data receiving step of receiving teacher data and a machine learning step of performing machine learning using the teacher data, wherein the teacher data receiving step is obtained by an analyzer different from the exhaust gas analyzer. A reference value for the specific component concentration, which is at least one of H ₂ concentration or O ₂ concentration, spectral data obtained by irradiating the combustion exhaust gas with light, or an element balance formula for obtaining the specific component concentration receiving teacher data containing at least one of the individual component concentrations selected based on the element balance formula or the calculated value of the specific component concentration calculated using the individual component concentrations in the element balance formula, and the machine learning step, The specific component correlation data is generated by machine-learning a relationship between the reference value of the specific component concentration and at least one of the spectral data, the individual component concentration, or the calculated value of the specific component concentration. .

In addition, the machine learning program according to the present invention is a machine used in an exhaust gas analyzer that irradiates light on combustion exhaust gas, detects light that has passed through the combustion exhaust gas, and analyzes the combustion exhaust gas based on the detection signal. A learning program, wherein a computer is provided with a function as a training data reception unit that receives training data and a function as a machine learning unit that performs machine learning using the training data, and the training data reception unit receives the exhaust gas A reference value for the specific component concentration, which is at least one of H ₂ concentration or O ₂ concentration, obtained by an analyzer different from the analysis device, spectral data obtained by irradiating the combustion exhaust gas with light, or the specific A teacher containing at least one of individual component concentrations selected based on an element balance formula for obtaining component concentrations, or a calculated value of a specific component concentration calculated using the individual component concentrations in the element balance formula The machine learning unit machine-learns the relationship between the reference value of the concentration of the specific component and at least one of the spectral data, the concentration of the individual component, or the calculated value of the concentration of the specific component. It is characterized in that the specific component correlation data is generated by

Furthermore, the exhaust gas analysis method according to the present invention is a method for analyzing combustion exhaust gas using a light source for irradiating the combustion exhaust gas with light and a photodetector for detecting light transmitted through the combustion exhaust gas. , a specific component concentration that is at least one of H ₂ concentration or O ₂ concentration in the flue gas, spectrum data obtained by irradiating the flue gas with light, or an element balance for obtaining the specific component concentration Using specific component correlation data obtained by learning a relationship between at least one of individual component concentrations selected based on the formula and calculated values of specific component concentrations calculated using the individual component concentrations in the element balance formula , the specific component concentration in the flue gas is calculated from at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration, and the specific component correlation data.

In addition, an exhaust gas analysis program according to the present invention is an exhaust gas analysis program used in an exhaust gas analyzer using a light source for irradiating light on combustion exhaust gas and a photodetector for detecting light transmitted through the combustion exhaust gas. A specific component concentration that is at least one of H ₂ concentration or O ₂ concentration in the flue gas, spectral data obtained by irradiating the flue gas with light, or obtaining the specific component concentration Specific component correlation data obtained by learning a relationship between at least one of individual component concentrations selected based on an element balance formula, or a calculated value of a specific component concentration calculated using the individual component concentrations in the element balance formula. function as a specific component correlation data storage unit for storing, at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration, and the specific component correlation data, the specific component in the combustion exhaust gas The computer is provided with a function as a specific component concentration calculation unit that calculates the concentration.

According to the present invention described above, it becomes possible to measure the H ₂ concentration or the O ₂ concentration that had to be measured using another analyzer in the exhaust gas analyzer.

1 is an overall view of an exhaust gas measurement system including an exhaust gas analyzer according to one embodiment of the present invention; FIG. It is a schematic diagram which shows the whole exhaust gas analyzer in the same embodiment. It is a basic functional block diagram of the arithmetic processing unit in the same embodiment. It is a functional block diagram of the machine learning device in the same embodiment. It is a functional block diagram of the arithmetic processing unit in the same embodiment. FIG. 11 is a functional block diagram of a machine learning device in a modified embodiment; FIG. 11 is a functional block diagram of a machine learning device in a modified embodiment; It is a functional block diagram of an arithmetic processing unit in a modified embodiment. FIG. 11 is a functional block diagram of a machine learning device in a modified embodiment; It is a functional block diagram of an arithmetic processing unit in a modified embodiment. It is a functional block diagram of an arithmetic processing unit in a modified embodiment. FIG. 11 is a functional block diagram of a machine learning device in a modified embodiment;

An exhaust gas analyzer according to one embodiment of the present invention will be described below with reference to the drawings. It should be noted that all of the drawings shown below are schematically drawn with appropriate omissions or exaggerations for the sake of clarity. The same constituent elements are denoted by the same reference numerals, and descriptions thereof are omitted as appropriate.

The exhaust gas analyzer 100 of this embodiment constitutes a part of the exhaust gas measurement system 200, for example. As shown in FIG. 1, this exhaust gas measurement system 200 includes a chassis dynamo 300 and exhaust gas sampling for sampling combustion exhaust gas (hereinafter simply referred to as "exhaust gas") of a vehicle V, which is a test vehicle running on the chassis dynamo 300. It comprises an apparatus 400 and an analysis apparatus 100 that analyzes the measurement target component in the sampled exhaust gas.

<Basic Configuration of Exhaust Gas Analyzer 100>
Specifically, as shown in FIG. 2, the exhaust gas analyzer 100 is equipped with an infrared light source 1, an interferometer (spectroscopic unit) 2, a measurement cell 3, a photodetector 4, an arithmetic processing unit 5, and the like. It is an infrared gas analyzer using infrared spectroscopy (FTIR).

The infrared light source 1 emits infrared light having a broad spectrum (continuous light including light of many wavenumbers), and uses, for example, a tungsten/iodine lamp or a high-brightness ceramic light source.

The interferometer 2 uses a so-called Michelson interferometer, which includes a half mirror (beam splitter) 21, a fixed mirror 22 and a movable mirror 23, as shown in the figure. Light from the infrared light source 1 incident on the interferometer 2 is split by a half mirror 21 into reflected light and transmitted light. One light is reflected by the fixed mirror 22 , the other is reflected by the movable mirror 23 , returns to the half mirror 21 again, is synthesized, and exits from the interferometer 2 .

The measurement cell 3 is a transparent cell into which the sampled exhaust gas is introduced.

The photodetector 4 detects infrared light that has passed through the exhaust gas and outputs the detection signal (light intensity signal) to the arithmetic processing device 5 . The photodetector 4 of this embodiment is, for example, an MCT (HgCdTe) detector, but may be a photodetector having other infrared detection elements.

The arithmetic processing unit 5 has, for example, an analog electric circuit having a buffer, an amplifier, etc., a digital electric circuit having a CPU, a memory, a DSP, etc., and an A/D converter interposed therebetween. be.

The arithmetic processing unit 5 functions as a main analysis unit 51 as shown in FIG.

The main analysis unit 51 calculates transmitted light spectrum data representing the spectrum of light transmitted through the exhaust gas from the detection signal (light intensity signal) of the photodetector 4, and calculates infrared absorption spectrum data from the transmitted light spectrum data. , to identify various components in the exhaust gas and to calculate the concentration of each component.

The main analysis unit 51 includes a spectral data generation unit 511 and an individual component analysis unit 512.

When the movable mirror 23 is advanced and retreated and the intensity of the light transmitted through the exhaust gas is observed with the position of the movable mirror 23 as the horizontal axis, in the case of light with a single wave number, the light intensity draws a sine curve due to interference. On the other hand, since the actual light that has passed through the exhaust gas is continuous light, the sine curve differs for each wavenumber, and the actual light intensity is a superposition of the sine curves drawn by each wavenumber, and the interference pattern (interferogram) is form a wave packet.

The spectrum data generation unit 511 obtains the position of the movable mirror 23 by a rangefinder (not shown) such as a HeNe laser (not shown), and obtains the light intensity at each position of the movable mirror 23 by the photodetector 4. Fast Fourier transform (FFT) is applied to the obtained interference pattern to convert it into transmitted light spectrum data with each wave number component as the horizontal axis. Then, the transmitted light spectrum data of the exhaust gas is further converted into absorption spectrum data based on the transmitted light spectrum data previously measured with the measurement cell 3 empty, for example.

For example, the individual component analysis unit 512 determines various components (eg, CO, CO ₂ , NO, H ₂ O, NO ₂ , or hydrocarbon component (HC), etc.) is specified, the concentration of each component is calculated, and this is output as individual component concentration data.

<Machine learning device 6>
Next, the machine learning device 6 for measuring the H ₂ concentration or the O ₂ concentration in the exhaust gas using the exhaust gas analyzer 100 will be described.

The machine learning device 6 of the present embodiment performs machine learning by utilizing the fact that the H ₂ concentration and the O ₂ concentration can be estimated using the element balance formula obtained from the fuel combustion formula shown below. From the following elemental balance formula (conservation law of substance amount), the _H2 concentration can be linearly regressed with the concentrations of the components ( _CO2 , CO, _H2O , THC), and the _O2 concentration can be linearly regressed with the concentrations of the components ( _CO2 , CO , H ₂ O, THC, NO) can be linearly regressed. Moreover, since the H ₂ concentration and the O ₂ concentration can be estimated from the individual component concentrations, they can also be estimated from the spectral data for obtaining these individual component concentrations.

Here, at least one of CO ₂ concentration, CO concentration, H ₂ O concentration, or THC concentration can be used as the individual component concentration when calculating the H ₂ concentration. At least one of CO ₂ concentration, CO concentration, H ₂ O concentration, THC concentration, or NO concentration can be used as the individual component concentration when calculating the O ₂ concentration.

(Fuel combustion type)
_{CaHbOc + dO2} + _{eN2 + fCO2} + _gH2O
→ _{n1CO2 + n2CO} + _{n3H2O + n4H2 + n5O2} + _{n6NO + n7N2 + n8Ca'Hb' + r}
In the above formula, total hydrocarbons (THC) are represented by _{Ca'Hb '} , where a' and b' are the averaged numbers of C and H in each hydrocarbon, respectively.
r is another component, and since it is a very small amount compared to the other components, it can be ignored in calculating the H ₂ concentration and the O ₂ concentration.

(element balance formula)
C: a+f= _n1 + _n2 + _a'n8
H: b+2g=2n ₃ +2n ₄ +b'n ₈
O: c+2d+2f+g= _2n1 + _n2 + _n3 + _2n5 + _n6
N: 2e=n ₆ +2n ₇
n ₀ : total amount of substances in flue gas n ₀ =n ₁ +n ₂ +n ₃ +n ₄ +n ₅ +n ₆ +n ₇ +n ₈
x _k : molar fraction of component k (x _k = n _k /n ₀ ) (k = 1 to 8)
_CO2 , 2; CO, 3; _H2O , 4; _H2 ,
5; O ₂ , 6; NO, 7; N ₂ , 8; THC (C _a′ H _b′ )
Let a′n ₈ =n ₀ x _THC and treat in units of x _THC /ppmC.
u _k : molar fraction of intake air component j excluding fuel (u _j =j/(d+e+f+g))
j=d; _O2 , e; _N2 , f; _CO2 , g; _H2O
n _r : Total molar ratio of intake and exhaust excluding fuel n _r = (d + e + f + g) / n ₀

Then, the following relational expression is obtained from the elemental balance formula of C.

In addition, the following relational expression can be obtained from the elemental balance formula for C and H.

The above formula gives the formula for _H2 concentration shown below. Since the formula becomes complicated, CO ₂ in the intake air is ignored in the following formula because it is a minute amount compared to other components (u _CO2 =0).

Thus, the _H2 concentration can be linearly regressed with the concentrations of the components ( _CO2 , CO, _H2O , THC).

On the other hand, the following relational expression is obtained from the elemental balance formula of O and N. Since the formula becomes complicated, the fuel does not contain O, and CO ₂ and H ₂ O in the intake air are neglected because they are minute amounts compared to other components.

Since the ratio of the amount of O and N in dry air is constant, the above is a constant value (constant A). This gives the following formula:

Eliminating the _H2 term in the above equation for H ₂ concentration (H/C elemental balance) yields the following O ₂ concentration equation.

Thus, the _O2 concentration can be linearly regressed with the concentrations of the components ( _CO2 , CO, _H2O , THC, NO).

This machine learning device 6 is a computer having a CPU, a memory, an input/output interface, an AD converter, or input means such as a keyboard, etc., and the CPU and its peripheral devices cooperate according to the machine learning program stored in the memory. As a result, as shown in FIG. 4, the functions of a teacher data receiving unit 61 that receives teacher data and a machine learning unit 62 that performs machine learning using the teacher data are exhibited. The machine learning device 6 may be incorporated in the arithmetic processing device 5 of the exhaust gas analyzer 100 described above, or a part of the functions of the machine learning device 6 may be provided in the arithmetic processing device 5. Also good.

The teaching data receiving unit 61 receives the H ₂ concentration reference value obtained by an H ₂ analyzer (not shown) different from the infrared gas analyzer (exhaust gas analyzer) and the infrared gas analyzer (exhaust gas analyzer). Different teacher data including a reference value of O ₂ concentration obtained by an O ₂ analyzer (not shown) and spectral data obtained by an infrared gas analyzer are accepted. The spectral data included in this teacher data is the absorption spectral data generated by the spectral data generating section 511 of the arithmetic processing unit 5, but may be transmitted light spectral data of the exhaust gas. The H ₂ spectrometer may be, for example, a thermal conductivity gas spectrometer (TCD) or a mass spectrometer. Also, for the O ₂ analyzer, for example, a zirconia sensor, a magnetic oxygen concentration meter, or the like may be used.

_The machine learning unit 62 machine-learns the relationship between the H ₂ concentration reference value and the spectrum data, and performs H ₂ correlation data (a machine learning model for calculating H ₂ concentration) that indicates the correlation between the H 2 concentration and the spectrum data. _and the _O2 correlation data ( _{O2 concentration calculation} and an O ₂ correlation data generator 622 that generates a machine learning model for

Here, the _H2 correlation data (machine learning model for calculating the _H2 concentration) calculated by the H2 correlation data generation unit 621 is stored in _{the H2} correlation data storage unit 623, and is stored in the _O2 correlation data generation unit 622. The calculated O ₂ correlation data (machine learning model for calculating O ₂ concentration) is stored in the O ₂ correlation data storage unit 624 .

<Characteristic Configuration of Exhaust Gas Analyzer 100 (Measurement of H ₂ Concentration or O ₂ Concentration)>
Then, as shown in FIG. 5, the exhaust gas analyzer 100 can calculate the H 2 concentration using the H ₂ correlation data (machine learning model for calculating _{the H 2 concentration} ) generated by the machine learning device 6. _O2 concentration can be calculated using _O2 correlation data (machine learning model for calculating _O2 concentration).

Specifically, the arithmetic processing unit 5 of the exhaust gas analyzer 100 includes an H ₂ concentration calculation unit 52 that calculates the H ₂ concentration using the H ₂ correlation data, and an O ₂ concentration calculation unit 52 that calculates the O ₂ concentration using the O 2 correlation data _. and a density calculator 53 . The H ₂ correlation data is stored in the H ₂ correlation data storage unit 54 , and the O ₂ correlation data is stored in the O ₂ correlation data storage unit 55 .

Note that when part or all of the machine learning device 6 is incorporated in the arithmetic processing device 5, the _H2 correlation data storage unit 54 may be configured from the _H2 correlation data storage unit 623 of the machine learning device 6. Alternatively, the O ₂ correlation data storage unit 55 may be configured from the O ₂ correlation data storage unit 624 of the machine learning device 6 .

The H ₂ concentration calculator 52 calculates the H ₂ concentration in the exhaust gas from the spectrum data generated by the spectrum data generator 511 and the H ₂ correlation data.

Here, when the H ₂ correlation data is generated using the absorption spectrum data, the H ₂ concentration calculation unit 52 calculates the H ₂ concentration using the absorption spectrum data generated by the spectrum data generation unit 511. . Further, when the _H2 correlation data is generated using the transmitted light spectrum data, the _H2 concentration calculator 52 calculates the _H2 concentration using the transmitted light spectrum data generated by the spectrum data generator 511. do.

The O ₂ concentration calculator 53 calculates the O ₂ concentration in the combustion exhaust gas from the spectrum data generated by the spectrum data generator 511 and the O ₂ correlation data.

Here, when the O ₂ correlation data is generated using the absorption spectrum data, the O ₂ concentration calculation unit 53 calculates the O ₂ concentration using the absorption spectrum data generated by the spectrum data generation unit 511. . Further, when the O ₂ correlation data is generated using the transmitted light spectrum data, the O ₂ concentration calculation unit 53 calculates the O ₂ concentration using the transmitted light spectrum data generated by the spectrum data generation unit 511. do.

<Effects of this embodiment>
According to the analyzer 100 _{of the present embodiment configured in this way, the H 2 concentration or O 2 concentration in} the exhaust gas and _the photodetector 4 H ₂ concentration or O ₂ concentration is calculated from the spectral data obtained from the detection signal of the photodetector 4 using the correlation data (machine learning model) that learned the relationship with the spectral data obtained from the detection signal of become able to. As a result, in exhaust gas analysis using infrared light, the H ₂ concentration or O ₂ concentration that does not absorb infrared light can be measured.

<Modified Example of First Embodiment>
For example, as shown in FIG. 6, the teacher data receiving unit 61 obtains from the spectral data, in addition to the H ₂ concentration reference value, the O ₂ concentration reference value, and the spectral data as the teacher data, It may accept individual component concentrations selected based on an elemental balance equation. Then, in the machine learning unit 62, the H ₂ correlation data generation unit 621 machine-learns the relationship between the H ₂ concentration reference value, the spectrum data, and the individual component concentration, and calculates the H ₂ concentration, the spectrum data, and the individual component concentration. Generate H ₂ correlation data (machine learning model for calculating H ₂ concentration) showing the correlation of Further, the O ₂ correlation data generation unit 622 machine-learns the relationship between the reference value of the O ₂ concentration, the spectrum data, and the individual component concentration, and the O ₂ concentration indicating the correlation between the O 2 concentration, the spectrum data, and the individual component concentration _. Generate correlation data (machine learning model for _O2 concentration calculation). In this way, by including the individual component concentrations selected based on the element balance formula in addition to the spectral data in the teacher data, the measurement accuracy of the O ₂ concentration or H ₂ concentration can be improved.

<Modified Embodiment 1>
For example, as shown in FIG. 7, the teacher data receiving unit 61 stores the H ₂ concentration reference value obtained by an H ₂ analyzer different from the infrared gas analyzer (exhaust gas analyzer) and the infrared gas analyzer The standard value of _O2 concentration obtained by an _O2 analyzer different from the (exhaust gas analyzer), the spectral data obtained by the infrared gas analyzer (exhaust gas analyzer), and the individual components obtained from the spectral data It accepts teacher data including concentrations and The spectral data included in this teacher data may be transmitted light spectral data of the exhaust gas generated by the spectral data generating section 511, or may be absorption spectral data.

Here, the individual component concentration is the individual component concentration such as CO, CO ₂ , NO, H ₂ O, NO ₂ or hydrocarbon component (HC) analyzed by the individual component analysis unit 512 .

Then, the machine learning device 6 of this embodiment estimates the calculated value (estimated value) of the H ₂ concentration and the calculated value (estimated value) of the O ₂ concentration using the element balance formula. That is, from the above-mentioned elemental balance formula (conservation law of substance amount), the H ₂ concentration can be linearly regressed with the concentration of the component (CO ₂ , CO, H ₂ O, THC), and the O ₂ concentration can be linearly regressed from the component (CO ₂ , CO, H ₂ O, THC, and NO).

In the elemental balance formula described above, the THC concentrations a' and b' are unknown quantities, and the measured values of the individual components have not a little error. occurs. Therefore, in this embodiment, the minimum error value is obtained by minimizing the concentration error between the calculated value of the concentration of the specific component and the reference value, and the correlation between the minimum error value and the spectrum is calculated.

Specifically, the _machine learning unit 62 calculates the minimum H2 error value by minimizing the _H2 concentration error between _the reference value of the H2 concentration and the calculated value (estimated value) of the _H2 concentration calculated from the element balance formula. and a first _H2 correlation data generation unit 621a that generates first _H2 correlation data indicating the correlation between the minimum _H2 error value and the parameter used to calculate the minimum _H2 error value _; and a _2H2 correlation data generation unit 621b that calculates the relationship with the minimum value and generates _2H2 correlation data. Here, the parameters used for calculating the minimum H ₂ error value that minimizes the H ₂ concentration error are a′ and b′ representing the THC concentration in the elemental balance formula. Alternatively, the _H2 error minimum value may be calculated by calculating the minimization problem by adding a, b and/or intake air moisture of the fuel to the parameters.

In addition, the machine learning unit 62 calculates the minimum O2 error value by minimizing the _O2 concentration error between _the reference value of the _O2 concentration and the calculated value (estimated value) of the _O2 concentration calculated from the element balance formula. a first _O2 correlation data generator 622a for generating first _O2 correlation data indicating the correlation between the minimum O2 _error value and the parameter used to calculate the minimum _O2 error value _; and a second _O2 correlation data generation unit 622b that machine-learns the relationship with the value to generate the second _O2 correlation data. Here, the parameters used for calculating the minimum O ₂ error that minimizes the O ₂ concentration error are the THC concentrations a' and b' in the elemental balance equation. Alternatively, the _H2 error minimum value may be calculated by calculating the minimization problem by adding a, b and/or intake air moisture of the fuel to the parameters.

The _1H2 correlation data generated by the _1H2 correlation data generation unit 621a is data indicating the correlation between the "minimum _H2 error value" and the "parameter of the element balance formula used to calculate the minimum _H2 error value". is. The _2H2 correlation data generated by the _2H2 correlation data generating section 621b is data indicating the correlation between the "spectrum data" and the "minimum _H2 error value". Here, the _1H2 correlation data is stored in the _1H2 correlation data storage unit 623a, and the _2H2 correlation data is stored in the _2H2 correlation data storage unit 623b.

In addition, the first _O2 correlation data generated by the first _O2 correlation data generation unit 622a is the correlation between the "minimum _O2 error value" and the "parameter of the element balance formula used to calculate the minimum _O2 error value". data shown. The second _O2 correlation data generated by the second _O2 correlation data generating section 622b is data indicating the correlation between the "spectrum data" and the "minimum _O2 error value". Here, the first _O2 correlation data is stored in the first _O2 correlation data storage unit 624a, and the _{second O2} correlation data is stored in the second _O2 correlation data storage unit 624b.

Then, as shown in FIG. 8, the exhaust gas analyzer 100 uses the 1st _H2 correlation data and the 2nd _H2 correlation data (machine learning model for calculating the _H2 concentration) generated by the machine learning device 6 to calculate _H2. The concentration can be calculated, _and the O2 concentration can be calculated using the first _O2 correlation data and the second _O2 correlation data (machine learning model for calculating _O2 concentration).

Specifically, the arithmetic processing unit 5 of the exhaust gas analyzer 100 includes an _H2 error minimum value calculation unit 52a that calculates the _H2 error minimum value from the spectrum data and the second _H2 correlation data, and an _H2 error minimum value calculation unit and an _H2 concentration calculator 52b for calculating the _H2 concentration in the exhaust gas from the H2 error minimum value obtained by 52a and _the first _H2 correlation data. The _1H2 correlation data is stored in the _1H2 correlation data storage section 52c, and the _2H2 correlation data is stored in the _2H2 correlation data storage section 52d.

Further, the arithmetic processing unit 5 includes an O ₂ error minimum value calculation unit 53a that calculates the O ₂ error minimum value from the spectrum data and the second O ₂ correlation data, and an O ₂ error value obtained by the O 2 error minimum value calculation unit 53a. and an _O2 concentration calculator 53b for calculating the _O2 concentration in the exhaust gas from the 2nd error minimum value and _the first _O2 correlation data. The first _O2 correlation data is stored in the first _O2 correlation data storage section 53c, and the second _O2 correlation data is stored in the second _O2 correlation data storage section 53d.

Note that when part or all of the machine learning device 6 is incorporated in the arithmetic processing device 5, the first _H2 correlation data storage unit 52c or the second _H2 correlation data storage unit 52d, respectively, of the machine learning device 6 The machine learning device 6 may comprise the first _H2 correlation data storage unit 623a or the second _H2 correlation data storage unit 623b, respectively, or the first _O2 correlation data storage unit 53c or the second _O2 correlation data storage unit 53d, respectively. 1st _O2 correlation data storage section 624a or 2nd _O2 correlation data storage section 624b.

<Modified Example of Modified Embodiment 1>
In the modified embodiment 1, instead of the individual component concentrations contained in the teacher data, the calculated values of the specific component concentrations calculated from the individual component concentrations and the element balance formula may be used. Specifically, the teacher data may include the first correlation data indicating the minimum error value obtained by minimizing the error between the reference value of the concentration of the specific component and the calculated value of the concentration of the specific component. In this case, an information processing device (not shown) that generates the first correlation data is provided separately from the processing device 5, and the processing device 5 machine-learns the relationship between the spectrum data and the minimum error value. , a second correlation data generator for generating the second correlation data of the minimum error value for the spectrum.

<Modified Example of First Embodiment>
For example, as shown in FIG. 6, the teacher data receiving unit 61 obtains from the spectral data, in addition to the H ₂ concentration reference value, the O ₂ concentration reference value, and the spectral data as the teacher data, It may accept individual component concentrations selected based on an elemental balance equation. Then, in the machine learning unit 62, the H ₂ correlation data generation unit 621 machine-learns the relationship between the H ₂ concentration reference value, the spectrum data, and the individual component concentration, and calculates the H ₂ concentration, the spectrum data, and the individual component concentration. Generate H ₂ correlation data (machine learning model for calculating H ₂ concentration) showing the correlation of Further, the O ₂ correlation data generation unit 622 machine-learns the relationship between the reference value of the O ₂ concentration, the spectrum data, and the individual component concentration, and the O ₂ concentration indicating the correlation between the O 2 concentration, the spectrum data, and the individual component concentration _. Generate correlation data (machine learning model for _O2 concentration calculation). In this way, by including the individual component concentrations selected based on the element balance formula in addition to the spectral data in the teacher data, the measurement accuracy of the O ₂ concentration or H ₂ concentration can be improved.

<Modified Embodiment 2>
Further, as shown in FIG. 9, the teacher data receiving unit 61 stores the H ₂ concentration reference value obtained by an H ₂ analyzer different from the infrared gas analyzer (exhaust gas analyzer) and the infrared gas analyzer The teaching data including the reference value of the O ₂ concentration obtained by the O ₂ analyzer different from the (exhaust gas analyzer) and the individual component concentrations obtained by the infrared gas analyzer (exhaust gas analyzer) are accepted. The individual component concentrations included in this teacher data are individual component concentrations such as CO, CO ₂ , NO, H ₂ O, NO ₂ or hydrocarbon components (HC) analyzed by the individual component analysis unit 512 .

The machine learning unit 62 includes an H ₂ correlation data generation unit 621 for generating H ₂ correlation data by performing machine learning on the relationship between the reference value of the H ₂ concentration and the calculated value (estimated value) of the H ₂ concentration, and an O ₂ It has an O ₂ correlation data generation unit 622 for generating O ₂ correlation data by performing machine learning on the relationship between the reference value of concentration and the calculated value (estimated value) of O ₂ concentration.

Here, the calculated value (estimated value) of the H ₂ concentration and the calculated value (estimated value) of the O ₂ concentration can be estimated using the element balance formula obtained from the fuel combustion formula described above. That is, from the elemental balance formula (conservation law of substance amount), the H ₂ concentration can be linearly regressed with the concentrations of the components (CO ₂ , CO, H ₂ O, THC), and the O ₂ concentration can be linearly regressed from the components (CO ₂ , CO , H ₂ O, THC, NO) can be linearly regressed. Here, in the H ₂ concentration linear regression equation, by adding the NO concentration in addition to the concentrations of the components (CO ₂ , CO, H ₂ O, THC), the measurement accuracy of the H ₂ concentration can be improved. .

The _H2 correlation data (machine learning model for calculating _H2 concentration) generated by the H2 correlation data generation unit 621 is stored in _{the H2} correlation data storage unit 623, and the _O2 correlation data generation unit 622 generates The O ₂ correlation data (machine learning model for calculating O ₂ concentration) is stored in the O ₂ correlation data storage unit 624 .

Then, as shown in FIG _. 10, the exhaust gas analyzer 100 can calculate the H 2 concentration using the H ₂ correlation data (machine learning model for calculating the H ₂ concentration) generated by the machine learning device 6, _O2 concentration can be calculated using _O2 correlation data (machine learning model for calculating _O2 concentration).

Specifically, the arithmetic processing unit 5 of the exhaust gas analyzer 100 includes an H ₂ concentration calculation unit 52 that calculates the H ₂ concentration in the combustion exhaust gas from the individual component concentration and the H ₂ correlation data, and the individual component concentration and the O ₂ correlation data. and an O ₂ concentration calculator 53 for calculating the O ₂ concentration in the combustion exhaust gas from the data.

Note that when part or all of the machine learning device 6 is incorporated in the arithmetic processing device 5, the _H2 correlation data storage unit 54 may be configured from the _H2 correlation data storage unit 623 of the machine learning device 6. Alternatively, the O ₂ correlation data storage unit 55 may be configured from the O ₂ correlation data storage unit 624 of the machine learning device 6 .

Here, the H ₂ concentration calculation unit calculates a calculated value of the H ₂ concentration from the individual component concentrations, and calculates the H ₂ concentration in the flue gas from the calculated value and the H ₂ correlation data. Further, the O ₂ concentration calculation unit calculates a calculated value of O ₂ concentration from the individual component concentrations, and calculates the O ₂ concentration in the flue gas from the calculated value and the O ₂ correlation data.

　The THC concentration used in the above elemental balance formula is considered to be the THC concentration obtained by a THC analyzer different from the infrared gas analyzer (exhaust gas analyzer).

Further, as shown in FIG. 11, the exhaust gas analyzer 100 includes a THC correlation data storage unit 56 that stores THC correlation data obtained by learning the relationship between the THC concentration reference value and the spectrum data, and an infrared gas analyzer (spectrum The configuration further includes a THC concentration calculation unit 57 that calculates the THC concentration in the combustion exhaust gas from the spectral data and the THC correlation data obtained by the data generation unit 511), and the THC concentration calculation unit 57 obtains THC concentration may also be used.

Further, as shown in FIG. 12, in the machine learning device 6, the teacher data receiving unit 61 receives the reference value of the THC concentration obtained by a THC analyzer different from the infrared gas analyzer (exhaust gas analyzer) and the spectrum The machine learning unit 62 is configured to have a THC correlation data generation unit 625 that generates THC correlation data by machine learning the relationship between the THC concentration reference value and the spectrum data. Also good. The THC correlation data generated by this THC correlation data generation section 625 is stored in the THC correlation data storage section 626 .

The concentrations of individual components used in modified embodiments 1 and 2 above may be concentrations of all components CO ₂ , CO, H ₂ O, THC and NO, or CO ₂ , CO, H ₂ O and THC. , NO may be the concentration of some components.

In addition to performing machine learning using the calculated value of the concentration of the specific component calculated from the element balance formula, machine learning of the relationship between the concentration of the individual component and the concentration of the specific component is performed without obtaining the calculated value of the specific component concentration. It is good as

In addition, in Modified Embodiment 1, the _1H2 correlation data generation unit 621a or the _2H2 correlation data generation unit _621b generates individual component concentrations and/or O The correlation data may be calculated using the reference values of _two densities. Further, the first _O2 correlation data generation unit 622a or the second _O2 correlation data generation unit 622b uses the reference values of the individual component concentrations and/or _H2 concentrations in addition to the calculated _O2 concentration values obtained from the individual component concentrations. may be configured to calculate the correlation data. By increasing the parameters to be machine-learned in this way, it is possible to improve the measurement accuracy of the H ₂ concentration or the O ₂ concentration.

Furthermore, the exhaust gas measurement system of the above embodiment uses the chassis dynamo 300 to test the complete vehicle V. However, for example, an engine dynamometer may be used to test the performance of the engine. A dynamometer may be used to test powertrain performance.

The exhaust gas analyzer 100 may irradiate a measurement sample with light and analyze from the spectrum. As the exhaust gas analyzer 100, in addition to Fourier transform infrared spectroscopy, for example, NDIR, quantum cascade laser infrared spectroscopy, non-dispersive infrared absorption method, chemiluminescence method (chemiluminescence method), or these A combined method or the like may also be used. Moreover, the present invention is not limited to the analysis of exhaust gases from automobiles, and can also analyze exhaust gases emitted from internal combustion engines such as ships, aircraft, agricultural machinery, and machine tools, power plants, or incinerators. In addition to analyzing the tail pipe of an internal combustion engine or the flue of a power plant, the components of exhaust gas contained in the environment may also be analyzed. Further, the exhaust gas analyzer may use light other than infrared light.

In addition, various modifications and combinations of the embodiments may be made as long as they do not contradict the spirit of the present invention.

According to the above-described present invention, the H ₂ concentration or O ₂ concentration, which had to be measured using another analyzer, can be measured in the analyzer.

100 Exhaust gas analyzer 1 Infrared light source 4 Photodetector 5 Arithmetic processor 52 _H2 concentration calculator 52a _H2 concentration error calculator 52b 1st _H2 correlation data storage unit 52d 2nd _H2 correlation data _{storage unit 53 02 concentration calculation unit 53a 02 concentration error} calculation unit 53b First _O2 correlation data storage unit 53d Second _{O2 correlation} data storage unit 54 _H2 correlation data storage unit 55 _O2 correlation data storage unit 56 THC correlation data storage unit 57 THC concentration calculation unit 6 machine learning device 61 teacher data reception unit 62 machine learning unit 621 _H2 correlation data generation unit 622 _O2 correlation data generation unit 623 _H2 correlation data storage unit 624 _O2 correlation data storage unit 621a 1st _H2 correlation data generation unit 621b 2nd _H2 correlation data generation unit 622a 1st _O2 correlation data generation unit 622b 2nd _O2 correlation data generation unit 623a 1st _H2 correlation data storage unit 623b 2nd _H2 correlation data storage unit 624a 1st _O2 Correlation data storage unit 624b... _2nd O2 correlation data storage unit 625... THC correlation data generation unit 626... THC correlation data storage unit

Claims (15)

1. A machine learning device used in an exhaust gas analyzer for irradiating light on combustion exhaust gas, detecting light transmitted through the combustion exhaust gas, and analyzing the combustion exhaust gas based on the detection signal,
   a training data reception unit that receives training data;
   A machine learning unit that performs machine learning using the teacher data,
   The training data reception unit
   a reference value of specific component concentration, which is at least one of H ₂ concentration or O ₂ concentration, obtained by an analyzer different from the exhaust gas analyzer;
   Spectral data obtained by irradiating the combustion exhaust gas with light, or individual component concentrations selected based on the element balance formula for determining the specific component concentration, or using the individual component concentrations in the element balance formula and at least one of the calculated values of the concentration of the specific component calculated by
   The machine learning unit
   a reference value for the specific component concentration;
   A machine learning device for generating specific component correlation data by machine-learning a relationship with at least one of the spectrum data, the individual component concentration, or the specific component concentration calculated value.
2. The teacher data receiving unit receives teacher data including the reference value of the concentration of the specific component and the spectrum data,
   2. The machine learning device according to claim 1, wherein the machine learning unit machine-learns a relationship between the reference value of the concentration of the specific component and the spectral data to generate the specific component correlation data.
3. The training data reception unit further receives the concentration of the individual component as training data,
   3. The machine learning unit according to claim 2, wherein the machine learning unit generates the specific component correlation data by machine-learning the relationship between the reference value of the specific component concentration, the spectrum data, and the individual component concentration. Machine learning device.
4. The machine learning unit
   calculating the minimum error value by minimizing the error between the reference value of the concentration of the specific component and the calculated value of the concentration of the specific component, and calculating the correlation between the minimum error value and the parameter used to calculate the minimum error value; a first correlation data generation unit that generates the first correlation data shown as part of the specific component correlation data;
   Second correlation data for generating, as part of the specific component correlation data, second correlation data indicating correlation between the spectral data and the minimum error value by performing machine learning on the relationship between the spectral data and the minimum error value 4. The machine learning device according to any one of claims 1 to 3, further comprising a generator.
5. The teacher data receiving unit receives teacher data including the reference value of the concentration of the specific component and the concentration of the individual component,
   2. The machine learning device according to claim 1, wherein the machine learning unit generates the specific component correlation data by machine-learning a relationship between the reference value of the specific component concentration and the individual component concentration.
6. When machine learning is performed on H ₂ correlation data as the specific component correlation data, the individual component concentration is at least one of CO ₂ concentration, CO concentration, H ₂ O concentration, or THC concentration,
   In the case of machine learning O ₂ correlation data as the specific component correlation data, the individual component concentration is at least one of CO ₂ concentration, CO concentration, H ₂ O concentration, THC concentration, or NO concentration. Item 6. The machine learning device according to any one of Items 1 to 5.
7. The teacher data receiving unit receives teacher data including a reference value of THC concentration obtained by an analyzer different from the exhaust gas analyzer and the spectrum data,
   The machine learning device according to any one of claims 1 to 6, wherein the machine learning unit machine-learns a relationship between the THC concentration reference value and the spectrum data to generate THC correlation data.
8. 　The machine learning device according to claim 7, wherein the individual component concentrations include THC concentrations, and the THC concentrations are obtained from the spectral data and the THC correlation data.
9. An exhaust gas analyzer for analyzing combustion exhaust gas,
   a light source that irradiates the combustion exhaust gas with light;
   a photodetector that detects light transmitted through the combustion exhaust gas;
   a specific component correlation data storage unit for storing specific component correlation data generated by the machine learning device according to any one of claims 1 to 8;
   a specific component concentration calculation unit that calculates the specific component concentration in the combustion exhaust gas from at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration, and the specific component correlation data. , flue gas analyzer.
10. The exhaust gas analyzer according to claim 9, wherein the combustion exhaust gas is automobile exhaust gas.
11. The exhaust gas analyzer according to claim 9 or 10, which uses Fourier transform infrared spectroscopy.
12. A machine learning method used in an exhaust gas analyzer that irradiates light on a combustion exhaust gas, detects light transmitted through the combustion exhaust gas, and analyzes the combustion exhaust gas based on the detection signal,
    a teaching data receiving step for receiving teaching data;
    A machine learning step of performing machine learning using the teacher data,
    The teacher data receiving step includes:
    a reference value of specific component concentration, which is at least one of H ₂ concentration or O ₂ concentration, obtained by an analyzer different from the exhaust gas analyzer;
    Spectral data obtained by irradiating the combustion exhaust gas with light, or individual component concentrations selected based on the element balance formula for determining the specific component concentration, or using the individual component concentrations in the element balance formula receiving teacher data containing at least one of the calculated values of the specific component concentration calculated by
    The machine learning step includes:
    a reference value for the specific component concentration;
    A machine learning method for generating specific component correlation data by machine-learning a relationship with at least one of the spectrum data, the individual component concentration, or the specific component concentration calculated value.
13. A machine learning program used in an exhaust gas analyzer that irradiates light on combustion exhaust gas, detects light transmitted through the combustion exhaust gas, and analyzes the combustion exhaust gas based on the detection signal,
    a function as a training data reception unit that receives training data;
    Equipping a computer with a function as a machine learning unit that performs machine learning using the teacher data,
    The training data reception unit
    a reference value of specific component concentration, which is at least one of H ₂ concentration or O ₂ concentration, obtained by an analyzer different from the exhaust gas analyzer;
    Spectral data obtained by irradiating the combustion exhaust gas with light, or individual component concentrations selected based on the element balance formula for determining the specific component concentration, or using the individual component concentrations in the element balance formula and at least one of the calculated values of the concentration of the specific component calculated by
    The machine learning unit generates specific component correlation data by machine-learning a relationship between the reference value of the specific component concentration and at least one of the spectral data, the individual component concentration, or the calculated value of the specific component concentration. A machine learning program that
14. A flue gas analysis method for analyzing flue gas using a light source for irradiating the flue gas with light and a photodetector for detecting light transmitted through the flue gas,
    Using the specific component correlation data generated by the machine learning device according to any one of claims 1 to 8, at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration; A method for analyzing exhaust gas, wherein the specific component concentration in the combustion exhaust gas is calculated from the specific component correlation data.
15. An exhaust gas analysis program used in an exhaust gas analyzer using a light source for irradiating light on combustion exhaust gas and a photodetector for detecting light transmitted through the combustion exhaust gas,
    A function as a specific component correlation data storage unit that stores specific component correlation data generated by the machine learning device according to any one of claims 1 to 8;
    A function as a specific component concentration calculation unit that calculates the specific component concentration in the combustion exhaust gas from at least one of the spectrum data, the individual component concentration, or the calculated value of the specific component concentration, and the specific component correlation data; An exhaust gas analysis program that provides a computer with

Images